Turn a Photo into an Oil Painting

How it works

Your photo is scaled so its total area is around one megapixel, keeping the original proportions. An image-editing model reads that copy alongside an instruction describing a richly textured oil painting: visible brushwork, layered color, natural light, a coherent painterly background, with the subject, pose, and composition preserved.

A second instruction rules out distorted identity, malformed anatomy, added objects, unreadable text, and watermarks.

The model then repaints the image at high strength, starting from your photograph and traveling a long way from it. Because each run begins from fresh noise, a second attempt gives you a different painting rather than the same one again — worth knowing, because the first result is not the only one available.

What oil paint actually is

Every convincing oil painting effect is solving the same problem, and it is worth naming.

Oil paint is thick. It holds the mark of the brush, dries slowly enough to be worked wet into wet, and builds up in layers that sit physically above the canvas. Look at a real painting from the side and you can see the surface is not flat. Move your head and the highlights move, because they are reflections off actual ridges rather than pale marks in a picture.

None of that can exist in a file. What a painted image has instead is the visual evidence of relief — a bright edge along one side of a stroke, a shadow along the other, the direction of the brush implied by how color breaks. The model has learned that evidence thoroughly, because it has seen an enormous number of photographs of paintings, which are themselves flat images of relief.

So the imitation is faithful to the thing it was trained on. It reproduces what a painting looks like in a photograph, which is very nearly all anyone ever sees.

Brushstroke scale

This is the practical consequence, and it catches people at the last step.

Brushwork is generated in pixels at the working size. A stroke that reads naturally in a one megapixel image is perhaps thirty pixels across. Relative to that canvas it is a brush mark.

Upscale to a poster and that stroke becomes two hundred pixels across. It is still the same shape, still with its implied highlight, but relative to the new canvas it is no longer a brush mark — it is a smear the width of a hand. The painting stops reading as a painting and starts reading as an enlargement of one, which is exactly what it is.

Real painters choose stroke size against the size of the work. A portrait painted at gallery scale has larger marks than a miniature. Here the scale was decided at one megapixel before you knew what you would do with the file.

The practical rule: use the output near the size it was made, and if you must print large, accept that the brushwork will read as coarse rather than fine. It is a look, but it is not the look you approved on screen.

Faithfulness at high strength

The repaint runs at high strength, which is what buys the brushwork.

A light pass leaves your photograph visible underneath and produces something closer to a texture overlay — a photo wearing paint. A heavy pass produces an actual painting, with invented light, simplified form, and color decisions the photograph did not contain.

The cost is that the same latitude applies to everything, including faces. The negative instruction names distorted identity first for that reason. It helps. It does not make the drift impossible, and on a portrait of someone specific, that drift is the only thing that matters.

Landscapes, buildings, and still life have no such exposure. Nobody can tell you a cloud is the wrong shape.

Publication gate

This page ships once the workflow has been run against a portrait, a landscape, and a subject containing small text, with likeness checked against the source and one output upscaled to print size to confirm how the stroke scale behaves.